---

name: simulink-simulation description: Use this skill whenever simulating a Simulink model, running the sim command, setting up SimulationInput objects, passing input signals via timeseries or datasets, configuring model or block parameters programmatically, or accessing logged output data from SimulationOutput. Trigger for any request involving sim(), Simulink.SimulationInput, Simulink.SimulationOutput, logsout, setExternalInput, or setModelParameter.

Simulating Simulink Models with the sim Command

Minimal working pattern

Always simulate using Simulink.SimulationInput and Simulink.SimulationOutput:

in = Simulink.SimulationInput('MyModel');
in = in.setModelParameter('StopTime', '10');
out = sim(in);

Never use set_param, load_system, or open_system to drive simulation — the SimulationInput API replaces all of these.

If the MATLAB MCP tool is unavailable but the user explicitly asks for real artifacts such as .slx, exported figures, or .mat results, use local MATLAB execution such as matlab -batch as a fallback. Give MATLAB generous timeouts because startup alone can take minutes.

For MATLAB MCP core server v0.9.0 and newer, this personal plugin launcher can pass through optional session/logging settings via environment variables:

• MATLAB_SESSION_MODE -> --matlab-session-mode
• MATLAB_EXTENSION_FILE -> --extension-file
• MATLAB_EXTENSION_FILES -> repeated --extension-file values, separated by the OS path separator (; on Windows)
• MATLAB_LOG_FOLDER -> --log-folder
• MATLAB_LOG_LEVEL -> --log-level

MATLAB MCP core server v0.9.2 improves --initialize-matlab-on-startup=true so the MCP server is not blocked while MATLAB starts, and fixes shareMATLABSession() failures on Windows. Keep the launcher default lazy startup unless the user explicitly wants startup-time initialization.

MATLAB MCP core server v0.10.0 makes auto the default session mode. Leave MATLAB_SESSION_MODE unset for normal use, set it to existing only when the user has run shareMATLABSession() in the target MATLAB session, and use multiple extension files only when the current MCP server exposes those custom tools.

Keep MATLAB_SETUP_MATLAB unset for normal MCP sessions; set it only for a one-time upstream setup run.

Setting parameters

Use SimulationInput methods to configure the simulation:

% Model-level parameters (StopTime, SolverType, SimulationMode, etc.)
in = in.setModelParameter('StopTime', '10', 'SolverType', 'Fixed-step');

% Block parameters
in = in.setBlockParameter('MyModel/Gain', 'Gain', '5');

% MATLAB workspace variables used by the model
in = in.setVariable('Kp', 1.2);

Input signals

Pass input signals through Inport blocks using a Simulink.SimulationData.Dataset. Each timeseries Name must match the corresponding Inport block's signal name, otherwise the signal won't be routed correctly.

dt = 0.01; % sample time
N = 1000; % Number of points
t = dt*(0:N)';
u = sin(2*pi*t);

ts = timeseries(u, t);
ts.Name = 'mySignal';   % must match the Inport signal name in the model

ds = Simulink.SimulationData.Dataset;
ds{1} = ts;

in = in.setExternalInput(ds);
out = sim(in);

Discovering logged signals

Before accessing logged data by name, discover what signals the model actually logs by running a simulation and inspecting logsout:

in = Simulink.SimulationInput('MyModel');
out = sim(in);
disp('List of logged signals:');
disp(out.logsout.getElementNames);

This is especially useful when working with an unfamiliar model — the names returned here are exactly the names to use when calling out.logsout.get(...).

Accessing logged data

Logged signals are available through out.logsout. Access them directly by name — no intermediate variables needed:

% Plot a logged signal
plot(out.logsout.get('signalName').Values)

% Get time and data separately
sig = out.logsout.get('signalName').Values;
plot(sig.Time, sig.Data)

Do not validate out with try-catch or isfield — sim either returns a valid SimulationOutput or throws an error. Simulink.SimulationOutput has no isfield method.

Accessing To Workspace Outputs

If the model uses To Workspace blocks and ReturnWorkspaceOutputs is on, retrieve those outputs directly from the SimulationOutput object:

in = Simulink.SimulationInput('MyModel');
in = in.setModelParameter('ReturnWorkspaceOutputs', 'on');
out = sim(in);

T = out.simout_T;
Q = out.simout_Qbase;

Normalize the exported value before post-processing because To Workspace may return a timeseries, a Simulink.SimulationData.Signal, or a numeric matrix depending on block settings:

if isa(T, 'timeseries')
    t = T.Time(:);
    y = T.Data(:);
elseif isa(T, 'Simulink.SimulationData.Signal')
    t = T.Values.Time(:);
    y = T.Values.Data(:);
else
    t = T(:,1);
    y = T(:,2);
end

Multiple simulations

When running many simulations, create an array of Simulink.SimulationInput objects using the repmat function instead of looping over sim:

in = repmat(Simulink.SimulationInput('MyModel'),N,1);
for k = 1:N
    in(k) = Simulink.SimulationInput('MyModel');
    in(k) = in(k).setVariable('gain', gains(k));
end
out = sim(in);

When collecting results from multiple cases into a struct array, preallocate a homogeneous struct shape before the loop. MATLAB will throw "subscripted assignment between dissimilar structures" if later cases add fields or nested layouts that the first element did not have.

Parallel simulation (parsim)

To run multiple simulations, use parsim instead of looping over sim:

for k = 1:N
    in(k) = Simulink.SimulationInput('MyModel');
    in(k) = in(k).setVariable('gain', gains(k));
end
out = parsim(in);

Rebuild-and-Simulate Pitfalls

If a script repeatedly rebuilds and reruns a model:

• Save or clear the dirty flag before close_system, otherwise MATLAB can warn that the changed model cannot be closed.
• Avoid model names that shadow scripts or functions on the MATLAB path; model/script name collisions can produce confusing warnings during sim and run.
• When a generated model relies on MATLAB Function blocks with workspace parameters, make sure those symbols are declared as block parameters during model construction; otherwise simulation can fail before sim starts with output-size/type inference errors.